Re: [syzbot] [rdma?] INFO: task hung in add_one_compat_dev (3)

Hi Jason,

On Tue, Oct 22, 2024 at 4:29 PM Jason Gunthorpe [off-list ref] wrote:

On Tue, Oct 22, 2024 at 12:39:27AM -0700, syzbot wrote:

quoted

1 lock held by syz-executor/27959:
 #0: ffffffff8fcbffc8 (rtnl_mutex){+.+.}-{3:3}, at: rtnl_lock net/core/rtnetlink.c:79 [inline]
 #0: ffffffff8fcbffc8 (rtnl_mutex){+.+.}-{3:3}, at: __rtnl_newlink net/core/rtnetlink.c:3749 [inline]
 #0: ffffffff8fcbffc8 (rtnl_mutex){+.+.}-{3:3}, at: rtnl_newlink+0xab7/0x20a0 net/core/rtnetlink.c:3772

There is really something wrong with the new sykzaller reporting, can
someone fix it?

The kernel log that shows the programs:

https://syzkaller.appspot.com/x/log.txt?x=10d72727980000

Doesn't have the word "newlink"/"new"/"link" etc, and yet there is an
executor clearly sitting in a newlink netlink callback when we
crashed.

These are likely coming from the network devices initialization code.
When syzbot spins up a new syz-executor, it creates a lot of
networking devices as one of the first steps.
https://github.com/google/syzkaller/blob/f00eed24f2a1332b07fef1a353a439133978d97b/executor/common_linux.h#L1482

So those syz-executors might have just been unable to start and then
they were abandoned (?)

We need to see the syzkaller programs that are triggering these issues
to get ideas, and for some reason they are missing now.

Once syzbot manages to find a reproducer, hopefully things will become
more clear.

-- 
Aleksandr

Jason